Scientists at New York University have shown that adults and even babies have an advantage over artificial intelligence in the ability to draw the right conclusions about what drives the actions of other people. This is reported in an article published in the journal Cognition.
To determine the fundamental differences between human and AI abilities, the researchers conducted experiments on 11-month-old infants using the previously established Boundary of Intuition for Children (BIB), six tasks that explore the psychology of common sense. The BIB was designed to test the intelligence of both infants and machine intelligence.
In particular, the babies watched a series of videos with simple animated figures. The actions of the figures imitated human behavior and decision making through the search for objects on the screen and other movements. The researchers also trained neural network models and tested their response to the same videos.
It turned out that the infants made correct predictions that the actions of the figures on the screen are due to hidden but constant goals – for example, searching for the same object no matter where it is. At the same time, children looked longer at objects whose behavior contradicted their predictions, a feature scientists called the “surprise paradigm.” This allowed the researchers to compare the measure of surprise in humans with that of algorithms, which failed to match infants in their ability to make intuitive predictions.